Industrial anomaly detection

September 5, 2026 · View on GitHub

Indad operator console with dataset gallery, readiness checks, and sample review

Build and inspect industrial anomaly datasets with a workflow designed for operators and automation: create → import → check → review → run a baseline. The focus is persistent datasets, actionable quality checks, and a small set of shared model abstractions.

Start the operator workspace:

python -m pip install -e ".[web]"
indad-web

Open http://127.0.0.1:8000.

Create datasets, save healthy and defective images, review masks and dataset health, then train a baseline. Export the same checks and sample inventory as JSON for agents:

indad-data inspect datasets/your_dataset --require inspection > manifest.json

See the operator and agent dataset workflow for creation, imports, readiness rules, and the Python API.

Three core anomaly detection baselines support this workflow:

SPADE (Cohen et al. 2021) - knn in z-space and distance to feature maps
PaDiM* (Defard et al. 2020) - distance to multivariate Gaussian of feature maps
PatchCore (Roth et al. 2021) - knn distance to avgpooled feature maps

* actually does not have any knn mechanism, but shares many things implementation-wise.

Current version: 0.4.0. Releases follow Semantic Versioning; see the changelog for release notes.

Install

python -m pip install -e .

Install optional UI and export support with python -m pip install -e ".[web,export]". Python 3.10 or newer is required.

Usage

CLI:

indad METHOD [--dataset DATASET]
# or: python run.py METHOD [--dataset DATASET]

Results can be found under ./results/.

Code example:

from indad import MVTecDataset, SPADE

model = SPADE(k=5, backbone_name="resnet18", device="cpu")

# get some training data
class_name = "bottle"
train_ds, test_ds = MVTecDataset(class_name).get_dataloaders()

model.fit(train_ds)  # dataloaders currently use batch_size=1

# evaluate
image_rocauc, pixel_rocauc = model.evaluate(test_ds)
print(image_rocauc, pixel_rocauc)

Fitted memory banks are included in the model state, so normal PyTorch persistence works. Recreate the model with the same constructor options before loading:

import torch

torch.save(model.state_dict(), "spade.pt")
restored = SPADE(k=5, backbone_name="resnet18", pretrained=False)
state = torch.load("spade.pt", map_location="cpu", weights_only=True)
restored.load_state_dict(state)

Custom datasets

Use the dataset workspace or indad-data create and indad-data import to build a persistent MVTec-compatible collection. See the dataset workflow. Visual inspection supports datasets without masks; CLI benchmark evaluation requires pixel-level ground truth.

Results

📝 = paper, 👇 = this repo

Image-level

class	SPADE 📝	SPADE 👇	PaDiM 📝	PaDiM 👇	PatchCore 📝	PatchCore 👇
bottle	-	98.8		99.8	■100.0■	■100.0■
cable	-	76.5		93.3	■99.5■	96.2
capsule	-	84.6		88.3	98.1	95.3
carpet	-	84.3		■99.4	98.7	98.7
grid	-	37.1		98.2	■98.2■	93.0
hazelnut	-	88.7		83.7	■100.0■	100.0
leather	-	97.1		99.9	■100.0■	100.0
metal_nut	-	74.6		99.4	■100.0■	98.3
pill	-	72.6		89.0	■96.6■	92.8
screw	-	53.1		83.0	98.1	96.7
tile	-	97.8		98.6	98.7	■99.0■
toothbrush	-	89.4		97.2	■100.0■	98.1
transistor	-	89.2		96.8	■100.0■	99.7
wood	-	98.3		98.9	■99.2■	98.8
zipper	-	96.7		89.5	■99.4■	98.4
averages	85.5	82.6	95.3*	94.3	■99.1■	97.7

PaDiM average referencing PaDiM-WR50-Rd550

Pixel-level

class	SPADE 📝	SPADE 👇	PaDiM 📝	PaDiM 👇	PatchCore 📝	PatchCore 👇
bottle	97.5	97.7	98.3	97.8	■98.6■	97.8
cable	93.7	94.3	96.7	96.1	■98.5■	97.4
capsule	97.6	98.6	98.5	98.3	■98.9■	98.3
carpet	87.4	99.0	99.1	98.6	■99.1■	98.3
grid	88.5	96.1	97.3	97.2	■98.7■	96.7
hazelnut	98.4	98.1	98.2	97.5	■98.7■	98.1
leather	97.2	99.2	99.2	98.7	■99.3■	98.4
metal_nut	■99.0■	96.1	97.2	96.5	98.4	96.2
pill	■99.1■	93.5	95.7	93.2	97.6	98.7
screw	98.1	98.9	98.5	97.8	■99.4■	98.4
tile	■96.5■	93.3	94.1	94.8	95.9	94.0
toothbrush	■98.9■	■98.9■	98.8	98.3	98.7	98.1
transistor	■97.9■	96.3	97.5	97.2	96.4	97.5
wood	94.1	94.4	94.7	93.6	■95.1■	91.9
zipper	96.5	98.2	98.5	97.4	■98.9■	97.6
averages	96.9	96.8	97.5	96.9	■98.1■	97.2

PatchCore-10 was used.

Hyperparams

The following parameters were used to calculate the results. They more or less correspond to the parameters used in the papers.

spade:
  backbone: wide_resnet50_2
  k: 50
padim:
  backbone: wide_resnet50_2
  d_reduced: 350
  epsilon: 0.04
patchcore:
  backbone: wide_resnet50_2
  f_coreset: 0.1
  n_reweight: 3

Direction

Prioritize dataset creation, review, quality checks and reproducibility for operators and agents. Keep SPADE, PaDiM and PatchCore as the core models. The next dataset milestones are capture-group splits, versioned snapshots, and saved inspection runs with threshold review. See the workflow and roadmap.

Design considerations

Data is processed in single images to avoid batch-statistics interference. Models validate this contract and accept arbitrary spatial input sizes.
I decided to implement greedy kcenter from scratch and there is room for improvement.
torch.nn.AdaptiveAvgPool2d for feature map resizing, torch.nn.functional.interpolate for score map resizing.
GPU is used for backbones and, when available, coreset selection. Pass device="cpu" or --device cpu for deterministic CPU execution.
Historical coreset-selection performance:
- 400-500 it/s @ float32 (RTX3080)
- 1000+ it/s @ float16 (RTX3080)

Operator console

Install the web extra and run indad-web. Open http://127.0.0.1:8000. The responsive browser interface includes dataset creation, imports, a searchable sample gallery, mask review, label correction, reversible sample exclusion, quality findings, and background baseline inspection. Use indad-web --datasets /path/to/datasets --port 8000 to configure local storage.

The console is plain HTML/CSS/JavaScript served by FastAPI, with no frontend build step. Agents can use the same HTTP API; interactive documentation is at /docs and the OpenAPI schema at /openapi.json. See the HTTP API guide.

Development

python -m pip install -e ".[dev,web,export]"
make lint
make test

The default tests use a small deterministic backbone and do not download model weights. ONNX coverage is marked export; run it with pytest -m export when the export dependencies are installed.

Acknowledgements

hcw-00 for tipping sklearn.random_projection.SparseRandomProjection.
h1day for contributing the original heatmap range controls.
MVTec dataset from https://www.mvtec.com/company/research/datasets/mvtec-ad, please note that this data is CC BY-NC-SA 4.0.

References

SPADE:

@misc{cohen2021subimage,
      title={Sub-Image Anomaly Detection with Deep Pyramid Correspondences},
      author={Niv Cohen and Yedid Hoshen},
      year={2021},
      eprint={2005.02357},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

PaDiM:

@misc{defard2020padim,
      title={PaDiM: a Patch Distribution Modeling Framework for Anomaly Detection and Localization},
      author={Thomas Defard and Aleksandr Setkov and Angelique Loesch and Romaric Audigier},
      year={2020},
      eprint={2011.08785},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

PatchCore:

@misc{roth2021total,
      title={Towards Total Recall in Industrial Anomaly Detection},
      author={Karsten Roth and Latha Pemula and Joaquin Zepeda and Bernhard Schölkopf and Thomas Brox and Peter Gehler},
      year={2021},
      eprint={2106.08265},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

MVTec dataset:

@article{Bergmann2021,
 author = {Paul Bergmann and Kilian Batzner and Michael Fauser and David Sattlegger and Carsten Steger},
 title = {The MVTec Anomaly Detection Dataset: A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection},
 journal = {International Journal of Computer Vision},
 year = {2021},
 volume = {129},
 number = {4},
 pages = {1038-1059},
 doi = {10.1007/s11263-020-01400-4}
}

@inproceedings{Bergmann2019,
 author = {Paul Bergmann and Michael Fauser and David Sattlegger and Carsten Steger},
 title = {MVTec AD — A Comprehensive Real-World Dataset for Unsupervised Anomaly Detection},
 booktitle = {IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
 year = {2019},
 pages = {9584-9592},
 doi = {10.1109/CVPR.2019.00982}
}